Hub group generation for search and recommendation systems

Abstract

The invention relates to hub group generation for search and recommendation systems. Some aspects relate to techniques for generating a hub using a generative model and grouping items into hub groups using the hub. The hub is generated by obtaining item information for an item, generating a cue using the item information, and causing a generative model to generate text of the hub using the cue (including a hub name and hub description for each hub). A hub embedding for each hub is also generated. When an item is to be returned to a user device (e.g., as a recommended item or search result item), an item embedding of the item and a hub embedding of the hub are used to assign each item to a particular hub to generate hub groups, each hub group including a hub name, a hub description, and the assigned item. User interface information is provided to present at least a portion of the hub group on the user device.

Description

CROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application claims priority to U.S. Provisional Application No. 63 / 543,655, filed on October 11, 2023, which is incorporated herein by reference in its entirety. Technical Field

[0002] The present application relates to pivot group generation for search and recommendation systems. Background Art

[0003] The amount of information and content available on the Internet and / or stored on devices continues to grow exponentially. Given the huge amount of information, search and recommendation technologies have become an integral part of locating relevant information. For example, search engines facilitate the rapid search of items in electronic databases (such as, for example, databases of available items on e-commerce systems). The search engine receives a search query and provides search results for items in response to the search query. For a given search query, the search engine can process the search query, user data, contextual data, and / or other inputs to identify the items most relevant to a particular search. The search results for the identified items can be presented on the user device in several different forms, for example, on a search result user interface. Recommendation systems are related technologies for recommending items to users. Various recommendation algorithms have been developed to select recommended items based on, for example, seed items, user behavior information, and user preferences. Such recommendation systems have become ubiquitous and are essential for content discovery on various platforms (such as online retail, music, image search, and video streaming). Summary of the invention

[0004] Some aspects of the present technology relate in particular to using generative models to generate hub groups, etc., which organize search result items or recommended items around a topic to facilitate item discovery. According to some aspects, item information from sampled items is used to generate a prompt, which may also include information instructing the generative model to generate a hub based on the item information. The prompt is provided as an input to a generative model, which generates hubs based on the item information. Each hub is associated with a topic discovered by the generative model from the item information, and includes a hub name and a hub description for the topic. A hub embedding for each hub is also generated. In some aspects, the generative model generates a hub embedding for the hub based on the item embedding of the item associated with the hub. When a search result item or recommended item is selected for presentation on a user device, a hub group is generated to organize these items. Items are assigned to hubs to form hub groups using their item embeddings and hub embeddings (e.g., based on a similarity metric, such as cosine similarity). Each hub group includes a hub name, a hub description, and one or more items. The hub group is then provided for presentation on the user device (eg, on a search results page, item page, or other user interface).

[0005] This Summary is provided to introduce a selection of concepts in a simplified form that are further described below in the Detailed Description. This Summary is not intended to identify key features or essential features of the claimed subject matter, nor is it intended to be used as an aid in determining the scope of the claimed subject matter. BRIEF DESCRIPTION OF THE DRAWINGS

[0006] The present technology is described in detail below with reference to the accompanying drawings, in which: Figure 1 is a block diagram illustrating an exemplary system according to some embodiments of the present disclosure; Figure 2A is a block diagram illustrating an example process for generating a hub according to some embodiments of the present disclosure; Figure 2B is a block diagram illustrating an example process for generating a pivot embedding according to some embodiments of the present disclosure; Figure 2C is a block diagram illustrating an example process for grouping items to generate a hub group according to some embodiments of the present disclosure; Figure 3 is a diagram providing an example of prompts input to a generative model to generate a pivot according to some embodiments of the present disclosure; Figure 4 is a diagram providing an example of an output of a hub having been generated by a generative model according to some embodiments of the present disclosure; Figure 5 is a diagram providing an example user interface with a hub group according to some embodiments of the present disclosure; Figure 6 is a diagram providing another example user interface with a hub group according to some embodiments of the present disclosure; Figure 7 is a flow chart illustrating a method for generating and providing a hub group according to some embodiments of the present disclosure; Figure 8 is a flow chart illustrating a method for generating a hub using a generative model according to some embodiments of the present disclosure; and Fig. 9 is a flow chart illustrating a method for generating a pivot embedding according to some embodiments of the present disclosure; Fig.10 is a flow chart illustrating a method for grouping items to generate a hub group according to some embodiments of the present disclosure; Fig.11 is a block diagram of an exemplary computing environment suitable for use in embodiments of the present disclosure. DETAILED DESCRIPTION

[0007] Although search / recommendation systems are useful tools for locating items, the shortcomings of existing search and recommendation technologies often result in the consumption of unnecessary amounts of computing resources (e.g., I / O costs, network packet generation costs, throughput, memory consumption, etc.). For example, current search systems often fail to allow users to efficiently find items of interest. This may be the result of search engines' limited understanding of queries. Search systems often operate based on keyword-based queries and may not fully understand some queries, such as complex natural language queries submitted by some users. Although some search systems provide filtering functions to filter search results, the available filters may be insufficient or unintuitive. In the context of recommendations, the recommended items provided by the recommendation system are often too similar or too diverse to the seed items, and the distance between the recommended items and the seed items cannot be adjusted. Although some recommendation systems personalize recommended items for users, this personalization is not available in many cases, such as when a user navigates to an item page from a third-party search engine.

[0008] Due to the shortcomings of existing search and recommendation technologies, users often have to submit multiple queries before finding the desired items. For example, a user may issue a first query to a search engine that returns a set of search results for a given item. The user may browse the search results and select some search results to access the corresponding items. The selection of items results in retrieving items from various content sources. In addition, in some cases, applications that support these items are launched to present these items. In the context of recommendations, when the recommended items are insufficient, the user may select some items to view the item pages and discover that these items are not the items the user is looking for. As discussed above, this often results in the user turning to query-based searches, which may involve issuing a large number of queries in an attempt to identify relevant items.

[0009] These repeated inputs lead to increased consumption of computing resources, etc. For example, repeated user queries lead to packet generation costs that adversely affect computer network communications. Each time a user issues a query, the content or payload of the query is usually supplemented with header information or other metadata within the packet in TCP / IP and other protocol networks. Therefore, when this function is multiplied by all the inputs required to obtain the desired data, repeatedly generating this metadata and sending it through the computer network will incur throughput and latency costs. In some cases, these repeated inputs (e.g., repeated clicks, selections, or queries) increase storage device I / O (e.g., excessive physical read / write head movement on non-volatile disks) because each time a user enters unnecessary information (such as entering several queries), the computing system often has to contact the storage device to perform a read operation or a write operation, which is time-consuming and error-prone, and will eventually wear out components such as read / write heads. In addition, if a user repeatedly issues a query, it is costly because processing the query consumes a lot of computing resources. For example, for some search engines, a query execution plan may need to be calculated each time a query is issued, which requires the search system to find the query execution plan with the lowest cost to fully execute the query. This reduces throughput and increases network latency, and wastes valuable time.

[0010] In view of these deficiencies in existing search and recommendation technologies, aspects of the technology described herein improve the functionality of computers themselves by providing a solution that enables search / recommendation systems to utilize generative models to generate hub groups that group similar items around different topics. Typically, generative models are used to generate hubs that present different topics for grouping items. In addition, hub embeddings for the hubs are generated. When items are selected for presentation on a user device (e.g., as search result items or recommended items), the item embeddings and the hub embeddings of the items are used to assign the items to hubs to form hub groups, which are provided for presentation on the user device. Hub groups facilitate the search / recommendation process to provide items of interest more efficiently.

[0011] More specifically, according to some aspects of the technology described herein, a generative model is used to generate hubs based on item information of a collection of items. Each hub corresponds to a topic discovered by the generative model from the item information, and includes a hub name and a hub description generated by the generative model. The hub name can be an identifier of the topic, and the hub description can be a brief description of the topic. In order to generate the hub, the item information of the collection of items is used to create a prompt, which can also include information instructing the generative model how to generate the hub (including how to generate the hub name and hub description of the hub). The prompt is provided as input to the generative model, and the generative model generates text of the hub name and the hub description as output. In some aspects, the items are from a given item category so that the generative model generates hubs related to the item category. For example, a collection of items can be sampled from items corresponding to a shoe category so that the generative model generates hubs related to items in the shoe category.

[0012] Generate a hub embedding for each hub. In some aspects, the generative model generates the hub embedding based on the item embeddings of the items associated with each hub. This can include items in the set of items used to generate the hub and / or other sampled items.

[0013] When search result items or recommended items are selected for presentation on a user device (e.g., in response to a search query or when an item page is presented with similar items), the search / recommendation system uses hubs and hub embeddings to generate hub groups to organize the items provided to the user device. In some aspects, item embeddings are obtained for items to be returned to the user device. Based on the item embeddings and the hub embeddings (e.g., using a similarity metric such as cosine similarity), the items are assigned to hubs to generate hub groups. Each hub group includes a hub name, a hub description, and one or more items. In some aspects, items and hub groups can be ranked such that hub groups are ordered based on hub group rankings, and items are ordered within each hub group based on item rankings.

[0014] The technical aspects described herein provide multiple improvements to existing search / recommendation technologies. For example, computing resource consumption is improved relative to the prior art. Specifically, providing a hub group that groups items around a specific topic allows the system to more efficiently provide related items as search results or recommended items. Users can view hub names and hub descriptions to identify topics related to them and explore items associated with related topics. This eliminates (or at least reduces) repeated user queries, search result selections, and item presentations because users can more easily identify related items without constantly entering various search queries to access search results and / or constantly making item selections to obtain additional information about the items presented. Therefore, the technical aspects described herein reduce computing resource consumption, such as packet generation costs. For example, a user query (e.g., an HTTP request) will only need to traverse a computer network once (or a fewer number of times relative to the prior art). Specifically, for an initial user query, the content or payload of the user query is supplemented with header information or other metadata once within a packet in a TCP / IP and other protocol network. This packet for a user query is only sent once or less times over the network. Therefore, metadata is not repeatedly generated and packets are continuously sent over a computer network.

[0015] In a similar manner, the technical aspects described herein improve storage device or disk I / O and query execution functions because they only need to go to the disk once (or fewer times relative to existing search technologies). As described above, the deficiencies of existing search and recommendation technologies lead to repeated user queries, search result selections, and item presentations. This results in multiple traversals of the disk. In contrast, the aspects described herein reduce storage device I / O because the user only provides minimal input, so the computing system does not have to frequently contact the storage device to perform read or write operations. For example, the system can respond with search result items or recommended items that organize these items in a manner that satisfies a single user query (or fewer queries relative to the prior art) and / or a single round of user intent to provide recommendations. Therefore, there is not much wear on components such as read / write heads because the disk I / O is greatly reduced.

[0016] Various configurations also improve query execution resource savings. Specifically, for example, the search system computes query execution plans for fewer queries relative to existing search techniques. Because the technical aspects described herein do not have to repeatedly compute query execution plans, this increases throughput and reduces network latency because fewer user queries need to be executed, unlike existing search techniques.

[0017] Now referring to the accompanying drawings, Figure 1is a block diagram of an example system 100 for generating and providing hub groups to organize items presented to users according to an embodiment of the present disclosure. It should be understood that this and other arrangements described herein are set forth only as examples. Other arrangements and elements (e.g., machines, interfaces, functions, commands, and functional groupings) may be used to supplement or replace the arrangements and elements shown, and some elements may be omitted entirely. In addition, many of the elements described herein are functional entities that may be implemented as discrete or distributed components or in combination with other components, and implemented in any suitable combination and location. The various functions described herein that are performed by one or more entities may be performed by hardware, firmware, and / or software. For example, the various functions may be performed by a processor that executes instructions stored in a memory.

[0018] System 100 is an example of a suitable architecture for implementing some aspects of the present disclosure. System 100 includes, among other components not shown, a user device 102 and a search / recommendation system 104 . Figure 1 Each of the user device 102 and the search / recommendation system 104 shown may include one or more computer devices, such as those discussed below. Fig.11 The computing device 1100. Figure 1 As shown, the user device 102 and the search / recommendation system 104 can communicate via a network 106, which may include, but is not limited to, one or more local area networks (LANs) and / or wide area networks (WANs). Such networking environments are common in offices, enterprise-wide computer networks, intranets, and the Internet. It should be understood that within the scope of the present technology, any number of user devices and servers may be employed within the system 100. Each may include a single device or multiple devices collaborating in a distributed environment. For example, the search / recommendation system 104 may be provided by multiple server devices that jointly provide the functionality of the search / recommendation system 104 as described herein. In addition, other components not shown may also be included in the network environment.

[0019] The user device 102 may be a client device located on the client side of the operating environment 100, and the search / recommendation system 104 may be located on the server side of the operating environment 100. The search / recommendation system 104 may include server-side software designed to work in conjunction with the client-side software on the user device 102 to implement any combination of features and functions discussed in this disclosure. For example, the user device 102 may include an application 108 for interacting with the search / recommendation system 104. The application 108 may be, for example, a web browser or a dedicated application for providing functions such as those described herein. This division of the operating environment 100 is provided to illustrate an example of a suitable environment, and for each implementation, any combination of the user device 102 and the search / recommendation system 104 is not required to remain as separate entities. Although the operating environment 100 shows a configuration in a networked environment with a separate user device and a search / recommendation system, it should be understood that other configurations of the combined components may be employed. For example, in some configurations, the user device may also provide search / recommendation capabilities.

[0020] User device 102 may include any type of computing device that a user can use. For example, in one aspect, a user device may be Fig.11 The type of computing device 1100 described. By way of example and not limitation, the user device 102 may be embodied as a personal computer (PC), a laptop computer, a mobile or mobile device, a smart phone, a tablet computer, a smart watch, a wearable computer, a personal digital assistant (PDA), an MP3 player, a global positioning system (GPS) or device, a video player, a handheld communication device, a gaming device or system, an entertainment system, an in-vehicle computer system, an embedded system controller, a remote control, an appliance, a consumer electronic device, a workstation, or any combination of these described devices, or any other suitable device that can present notifications. A user may be associated with the user device 102 and may interact with the search / recommendation system 104 via the user device 102.

[0021] According to some aspects, the search / recommendation system 104 can be part of a listing platform. Examples of listing platforms include e-commerce platforms, where listed products or services are available for purchase by users of client devices when navigating to the platform. Other examples of listing platforms include rental platforms that list various rental items (e.g., equipment, tools, real estate, vehicles, contract labor) and media platforms that list digital content items (e.g., digital content for streaming / downloading).

[0022] The functionality of the listing platform includes providing an interface capable of presenting item listings of items to users of the listing platform. Item listings of items that can be sold / rented / consumed via the listing platform are stored by the item listing data storage device 118. Each item listing may include a description associated with the item, including one or more of a price in currency, reviews, an image of the item, shipping options, ratings, item condition, item size, item color, etc. In some aspects, the item is associated with one or more categories including meta categories and leaf categories. For example, meta categories are each divisible into subcategories (or branch categories), while leaf categories are not divisible.

[0023] At a high level, the search / recommendation system 104 employs a generative model to generate hubs, including hub names and hub descriptions. When providing items to a user device, such as the user device 102, the search / recommendation system 104 uses the hubs to group the items into a plurality of hub groups. This can be, for example, in the context of returning search results in response to a search query, providing recommendations (e.g., on a home page, product page, or other location on an e-commerce website), or otherwise providing information of a plurality of items to a user device.

[0024] like Figure 1 As shown, the search / recommendation system 104 includes a hub generation component 110, a hub embedding component 112, an item grouping component 114, a user interface component 116, and an item list data storage device 118. In addition, the components of the search / recommendation system 104 can be other components that provide additional functions beyond the features described herein. The search / recommendation system 104 can be implemented using one or more server devices, one or more platforms with corresponding application programming interfaces, cloud infrastructure, etc. Figure 1 In the configuration of FIG. 1 , the search / recommendation system 104 is displayed separately from the user device 102 , but it should be understood that in other configurations, some or all of the functionality of the search / recommendation system 104 may be provided on the user device 102 .

[0025] In one aspect, the functions performed by the components of the search / recommendation system 104 are associated with one or more personal assistant applications, services, or routines. Specifically, such applications, services, or routines can be run on one or more user devices, servers, can be distributed across one or more user devices and servers, or implemented in the cloud. In addition, in some aspects, these components of the search / recommendation system 104 can be distributed across a network (including one or more servers and client devices in the cloud), and / or can reside on a user device. In addition, these components, the functions performed by these components, or the services performed by these components can be implemented on an appropriate abstract layer (e.g., operating system layer, application layer, hardware layer, etc.) of a computing system. Alternatively or additionally, the functions of these components and / or the technical aspects described herein can be performed at least in part by one or more hardware logic components. For example, but not limited to, the exemplary types of hardware logic components that can be used include field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), etc. Additionally, while functionality is described herein with respect to specific components shown in the example system 100, it is contemplated that in some aspects the functionality of these components may be shared or distributed across other components.

[0026] The hub generation component 110 of the search / recommendation system 104 uses a generative model to generate hubs. Typically, the hub generation component 110 accesses information about multiple items, for example from an item list data storage device 118 (although the item information can be obtained from other sources). The item information is provided as input to a generative model, which generates multiple hubs based on the item information. Each hub includes a hub name and a hub description, which include text generated by the generated model. In some aspects, the item information forms part of a prompt that is provided as input to the generative model. The prompt can be generated to include one or more rules that instruct the generative model how to generate the hub name and hub description. In some aspects, the hub generation component 110 generates the hub offline and stores information about the hub for use in generating a hub group online when the item is returned to a user device (such as user device 102).

[0027] Figure 2A A block diagram is provided showing an example process 200 for generating a hub using a generative model, which example process 200 may be performed by, for example Figure 1 The central generation component 110 is executed. Figure 2A As shown, data is sampled from the item data storage device 202, and the sampled data 204 is used to generate prompts 206 for input to the generative model 208. In some aspects, the item data storage device 202 corresponds to Figure 1The item list data storage device 118 of the embodiment of the present invention can be a collection of information about items provided for the purpose of generating a hub. For example, the item data storage device 202 can store item information related to items available on a listing platform (such as a listing platform used by an e-commerce website). However, it should be understood that in other aspects, the item data storage device 202 can be another source of item information. For example, the item data storage device 202 can be a collection of information about items provided only for the purpose of generating a hub.

[0028] In some aspects of the techniques described herein, sampled data 204 includes item information for items from a given item category. For example, to generate a hub for a shoe category, information for items categorized as shoes in item data storage device 202 is accessed. In other aspects, sampled data 204 is independent of any item category. Item information for a given item in sampled data 204 may include, for example, a title, an item ID, a price, item attributes, and / or an item description. Item information may include structured text (e.g., name / value pairs) and / or unstructured text.

[0029] Prompt 206 includes text that includes item information from sampled data 204. Prompt 206 may also include information for instructing generative model 208 to generate a hub based on the item information. In some aspects, prompt 206 includes one or more rules about the generation of the hub, such as how the hub name and hub description should be constructed. For example, a rule may instruct generative model 208 to generate a short and rhyming hub name. As another example, a rule may instruct generative model 208 to generate a hub description that describes the subject of the hub within a specific number of words (e.g., 25 words or less). According to some aspects, prompt 206 is generated using prompt engineering. Prompt engineering is a process that involves designing and making effective prompts to obtain accurate, relevant, and useful desired responses from a generative model. The process may involve iteratively creating experimental prompts, providing these experimental prompts as input to a generative model, and revising these experimental prompts based on output from the generative model.

[0030] For example, Figure 3 An example of a prompt 300 that can be provided as input to a generative model according to some aspects is provided. Figure 3As shown, prompt 300 includes information 302 with specific rules for instructing the generative model how to generate a hub. In this example, information 302 includes rules on how to generate a hub name and a hub description. Information 302 also includes rules for grouping items using the generated hub and how to perform grouping. Prompt 300 also includes item information 304. Although for clarity, item information 304 in prompt 300 only includes information for a single item, it should be understood that prompts can generally include information for any number of items. In addition, although prompt 300 provides an example in which item information 304 includes structured text (i.e., name / value pairs), item information may include unstructured text (e.g., a general text description of an item).

[0031] Reference again Figure 2A , the generative model 208 may include a language model that includes a set of statistical or probabilistic functions that perform natural language processing (NLP) to understand, learn and / or generate human natural language content. For example, a language model may be a tool that determines the probability of a given sequence of words appearing in a sentence (e.g., via NSP or MLM) or a natural language sequence. In short, it may be a model that is trained to predict the next word in a sentence. When a language model is trained on a large amount of data, it is called an LLM. Some examples of LLMs are GOOGLE's BERT and OpenAI's GPT-3 and GPT-4. These models have the ability to range from writing simple papers to generating complex computer codes (all of which are performed with limited or even no supervision). Therefore, LLMs may include deep neural networks that are very large (billions to hundreds of billions of parameters) and understand, process and generate human natural language by training on a large amount of text. These models can predict future words in a sentence, allowing them to generate sentences in a manner similar to human speech and writing.

[0032] According to some aspects, the generative model 208 includes a neural network. As used herein, a neural network includes at least three operating layers, although a neural network may include more than three layers (i.e., a deep neural network). The three layers may include an input layer, a hidden layer, and an output layer. Each layer includes neurons. Different types of layers and networks connect neurons in different ways. Neurons have weights, activation functions, and outputs, the activation function defining the output of the neuron given an input (including weights). Weights are adjustable parameters that cause the network to produce the correct output.

[0033] In some configurations, the generative model 208 is a pre-trained model (e.g., GPT-4) that has not yet been fine-tuned. In other configurations, the generative model 208 is a model that is built and trained from scratch, or is a pre-trained model that has been fine-tuned. In this configuration, the generative model 208 can be trained or fine-tuned using training data. For example, the training data can include pairs of item information and predefined hub names and hub descriptions (e.g., manually generated hub names and hub descriptions), and the generative model 208 is trained to adapt to the training data. During training, the weights associated with each neuron can be updated. Initially, the generative model 208 can include random weight values ​​or pre-trained weight values ​​that are adjusted during training. On the one hand, the generative model 208 is trained using back propagation. The back propagation process includes a forward pass, a loss function, a backward pass, and a weight update. The process is repeated using training data. For example, each iteration can include: providing item information as input to the model, generating output by the model, comparing the model output to predefined hub names and hub descriptions paired with the input item information (e.g., calculating loss), and updating the model based on the comparison. The goal is to update the weights of each neuron (or other model component) so that the generative model produces useful hub names and hub descriptions. Once trained, the weights associated with a given neuron can remain fixed. Other data passed between neurons can change in response to a given input. Retraining the network with additional training data can update one or more weights in one or more neurons.

[0034] Based on the prompt 206 with the item information, the generative model 208 outputs a plurality of hubs 210. Each hub output by the generative model 208 includes a hub name and a hub description around different topics identified from the item information. In some aspects, the hubs 210 are generated according to the rules set forth by the prompt 206.

[0035] In some aspects, the output of generative model 208 includes not only hub names and hub descriptions for multiple hubs, but also associates one or more items identified in sample data 204 and included as part of prompt 206 with each hub. In other words, generative model 208 groups items from sample data 204 with hubs 210. Each item can be associated with one or more hubs 210.

[0036] Figure 4 An example of output 400 generated by a generative model is provided. Figure 4As shown, output 400 includes multiple hubs generated by the generative model. For each hub, the generative model has generated a hub name (such as hub name 402 for the first listed hub) and a hub description (such as hub description 404 for the first hub). In the current example, the generative model also groups with each hub the items that were included as part of the prompt (which was provided as input to the generative model). Therefore, output 400 includes item information (such as item information 406 for the first hub) that identifies the items that the generative model has associated with each hub.

[0037] In some configurations (such as Figure 2A In the configuration shown in the figure, post-processing 212 is performed on the hubs 210 generated by the generative model 208. By way of example and not limitation, the post-processing 212 includes deduplication of the hubs 210. Specifically, similar hubs can be identified, and similar hubs can be combined and / or one or more hubs can be removed. In some configurations, the post-processing 212 can include a manual review of the generated hubs 210 by an administrator / developer to confirm that the generated hubs 210 are accurate and helpful in grouping items. In some aspects, the generative model 208 can be used to refine the hub group 210.

[0038] After any post-processing 212, the hub information 214 is stored. The hub information 214 for each hub generated by the generative model 208 may include, for example, a category ID (in the case where an item from a particular item category is used to generate the hub), a hub ID, a hub name, and a hub description. Figure 2A Not shown, but where the generative model 208 also groups items from the sample data 204 with each hub, the hub's item information 214 may also include a series of item information (e.g., item IDs) for identifying items associated with the hub.

[0039] Reference again Figure 1 , the search / recommendation system 104 also includes a hub embedding component 112. Typically, the hub embedding component 112 generates a hub embedding for each hub generated by the hub generation component 110. In the context of natural language processing (NLP) and machine learning, an embedding refers to a digital representation of text, images, audio, video, or other content in a way that captures its semantics. In some aspects, the hub embedding component 110 generates the hub embeddings offline and stores the hub embeddings for use in generating hub groups online when an item is returned to a user device (such as the user device 102).

[0040] In some configurations, the hub embedding component 112 generates a hub embedding based on the hub name and the hub description. For example, to generate a hub embedding for a given hub, the hub embedding component 112 can provide the hub name and the hub description to a machine learning model, which generates the hub embedding based on the hub name and the hub description. In some configurations, in order to provide a more robust hub embedding, the hub embedding component 112 associates multiple items with each hub and generates a hub embedding based on the item embeddings of the items associated with each hub. The item embedding can be a text-based embedding (i.e., an embedding generated by providing text from item information to a machine learning model) and / or a multimodal embedding (i.e., an embedding generated by providing multiple types of inputs from item information (such as text and images) to a machine learning model). The items associated with each hub used to generate the hub embedding can, for example, include items used by the hub generation component 110 to generate the hub, other sampled items, and / or a combination thereof.

[0041] Figure 2B A block diagram is provided showing an example process 220 for generating a pivot embedding, which may be performed by, for example Figure 1 The central embedded component 112 is used to execute. Figure 2B The use of the generative model 228 to associate items with, for example, Figure 2A The process 200 described above is used to generate a hub for association. Figure 2B As shown, data is sampled from the item data storage device 222, and the sampled data 224 is used to generate prompts 226 for input to the generative model 228. In some aspects, the item data storage device 222 corresponds to Figure 1 Item list data storage device 222 may store item information related to items available on a listing platform (such as a listing platform used by an e-commerce website). However, it should be understood that in other aspects, item data storage device 222 may be another source of item information. For example, item data storage device 222 may be a collection of information about items provided solely for the purpose of generating a pivot embed.

[0042] In some aspects of the techniques described herein where a hub is associated with an item category, sampled data 224 includes item information for items from the item category. For example, if the hub corresponds to a shoe category, information for items classified as shoes in item data storage device 222 will be accessed. In other aspects, sampled data 224 is independent of any item category. Item information for a given item in sampled data 224 may include, for example, a title, an item ID, a price, item attributes, and / or an item description. Item information may include structured text (e.g., name / value pairs) and / or unstructured text.

[0043] Prompt 226 includes text that includes item information from sampled data 224, hubs (including, for example, hub names and hub descriptions generated using process 200), and information for instructing generative model 228 to associate items identified by the item information with the hubs. In some aspects, prompt 226 can instruct generative model 228 to associate each item with only one hub; while in other aspects, prompt 226 can allow the generative model to associate each item with more than one hub. In some configurations, prompt 226 can also include items used by the hub generation process (e.g., process 200). In the event that the hub generation process also associates these items with the generated hubs, these associations can also be identified by the prompt, which can further assist generative model 228 in assigning items from sampled data 224 to hubs.

[0044] Given a prompt 226, a generative model 228 provides an output that assigns items from the sampled data 224 (as well as any items from the hub generation process included in the prompt) to hubs. As described above, in some aspects, each item is assigned to only one hub, while in other aspects, each item can be assigned to one or more hubs. This provides a hub map 230. The hub map 230 can associate a list of item IDs of the items assigned to each hub by the generative model 228 with each hub ID.

[0045] Item embeddings 232 for items mapped to the hub are obtained and used to generate a hub embedding 234 for the hub. Item embeddings 232 may, for example, include text-based embeddings generated based on text associated with each item or multimodal embeddings, such as embeddings generated based on text and images associated with each item. In some aspects, item embeddings 232 have been previously generated and stored, such that item embeddings 232 can be retrieved from a storage device. In other aspects, item embeddings 232 are generated as part of the hub embedding generation process 220.

[0046] The hub embedding for a given hub is generated based on the item embeddings of the items mapped to the hub in hub map 230. As an example and not limitation, the hub embedding for a hub may be generated as the average of the item embeddings of the items mapped to the hub.

[0047] After the hub embedding is generated, the hub embedding information is stored 236. The hub information 236 for each hub may include, for example, a category ID (in the case where an item from a particular item category is used to generate the hub), a hub ID, and a hub embedding.

[0048] Although for the purpose of generating pivot embeddings, Figure 2BAn example of sampling items from the item data storage device 222 and associating the items with a hub by the generative model 228 is provided, but other methods can be used to generate hub embeddings. For example, in some aspects, as part of the hub generation process, items used to generate hubs can be assigned to hubs, and hub embeddings can be generated based on the item embeddings of these items.

[0049] Turn again Figure 1 , the search / recommendation system 104 also includes an item grouping component 114. Typically, the item grouping component 114 uses hubs and hub embeddings to group items into hub groups based on item embeddings. Hub groups can be provided to a user device (such as user device 102) to organize items presented via the user device. The items to be grouped can be, for example: recommendations provided to the user device on an item page, homepage, or other location; or search results returned in response to a query from the user device. As described above, in some aspects, hubs (i.e., hub names and hub descriptions) and hub embeddings can be generated and stored offline. In this configuration, when an online operation of returning items to the user device is performed, the item grouping component 114 can use the stored information to group items. In other aspects, items can be assigned to hubs offline, and information associating items with hubs can be stored. In this configuration, when items are provided to the user device, the stored association between items and hub groups can be used to generate hub groups.

[0050] Figure 2C A block diagram is provided showing an example process 240 for grouping items within a hub to form a hub group, which example process 240 may be performed by, for example Figure 1 The item grouping component 114 is used to perform the above. Figure 2C As shown, item information 242 of a plurality of items is obtained. The items to be grouped may correspond to an item data storage device (e.g., Figure 1 The items may be items in an item data storage device 118 of an item store and may be identified in any of a variety of different ways within the scope of the embodiments described herein. By way of example only and not limitation, the items may include recommended items identified using any of a variety of different recommendation algorithms. For example, the items may be recommended items for display on an item page, a home page (e.g., the home page of an e-commerce website), or any other user interface area for presenting recommended items. In some cases, the recommended items may be personalized recommendations for a given user based on, for example, the user's browsing history, purchase history, and other user behavior. As another example, the items may include search result items identified in response to a user query.

[0051] An item embedding 244 is obtained for each item identified by the item information 242. The item embedding 244 may, for example, include a text-based embedding or a multimodal embedding generated based on the text associated with each item, such as an embedding generated based on the text and an image associated with each item. In some cases, item embeddings for the items have been previously generated and stored. In this case, the item information 242 may be an item ID that is used to retrieve the stored item embeddings. In other cases, item embeddings have not been previously generated and stored. In this case, the item information 242 may include text, images, or other content that is provided as input to a machine learning model to generate item embeddings as part of the item grouping process 240.

[0052] Given an item embedding 244 of an item identified by item information 242 and hub embeddings of various hubs (e.g., from Figure 2B In the case of hub embedding information 236 of the item information 242, a hub assignment 246 is generated in which each item is assigned to a hub. The hub assignment 246 may include a mapping of one or more item IDs to each hub ID in a plurality of hub IDs. The assignment of items to hubs may be based on a similarity metric between item embeddings and hub embeddings, such as cosine similarity. In some configurations, the grouped items correspond to a given item category, and the hubs used to determine the hub assignment 246 are based on that item category. For example, if all items identified by the item information 242 correspond to a shoe category, the hub embedding of the hub with the category ID of the shoe category is retrieved and used to generate the hub assignment 246. In other words, the hub for grouping items may be selected based at least in part on the item category of the grouped items. In another configuration, the hub assignment 246 is generated independent of the item category.

[0053] As shown in block 250, the hub assignment 246 is combined with information about the hub (such as the hub name and the hub description) to generate a hub group. Thus, each hub group includes a hub name, a hub description, and one or more items. In some configurations (such as Figure 2C In the configuration shown in the figure), the item ranker 248 also provides item ranking. The item ranking can be, for example, a ranking based on the relevance of the item to the seed item (for example, in the case of a recommendation on an item page, an item from the item page can serve as a seed item), the relevance to the user, and / or the relevance to the search query. The item ranking can be used to sort the items within each hub group. The item ranking can also be used to sort the hub groups with respect to each other. For example, a hub group ranking for each hub group can be generated based on the item ranking of each item within each hub group (for example, the average of the item rankings), and the hub groups can be sorted based on the hub group ranking.

[0054] The hub group information 252 is generated based on the combination of information. The hub group information 252 for each hub group may include, for example, a hub name, a hub description, and an ordered sequence of item IDs. In some aspects, a subset of hub groups and / or items to be provided to a user device may be selected based on hub group rankings and item rankings. For example, if only N hub groups are to be returned to a user device and there are initially more than N hub groups, the N hub groups with the highest hub group rankings may be selected. Similarly, if only M items are to be returned for each hub group and there are initially more than M items in the hub group, the M items with the highest item rankings may be selected. As shown in box 254, the hub group information is provided to the front end of the system for sending the hub group information to the user device to present the hub group on a user interface.

[0055] Reference again Figure 1 , the search / recommendation system 104 also includes a user interface component 116, which provides one or more user interfaces for interacting with the search / recommendation system 104. The user interface component 116 provides one or more user interfaces to a user device, such as the user device 102. In some cases, the user interface can be presented on the user device 102 via an application 108, which can be a web browser or a dedicated application for interacting with the content generation system 104. For example, the user interface component 116 can provide a user interface for interacting with the search / recommendation system 104 to submit a search query or other input to the search / recommendation system 104, etc. The user interface component 116 can also provide a user interface for presenting a hub group and allowing a user associated with the user device to interact with the hub group.

[0056] Figure 5 and Figure 6 An example of a user interface presenting a hub group generated according to some aspects of the present technology is provided. Figure 5, a user interface 500 is provided that presents a recommendation section labeled "Similar Items," which may be provided, for example, on an item page of a particular item list to present items similar to the items of the item page. The user interface 500 includes a hub tab area 502 that presents user interface elements (e.g., "buttons") that are selectable by a user, and hub names of available hub groups. Each button in the hub tab area 502 includes a user-selectable user interface element that, when selected, presents a hub description of the associated hub group and the items within the hub group. For example, a button 504 for a hub group having a hub name of "Mod and GoGo" has been selected. Thus, a hub description 506 for the "Mod and GoGo" hub group is presented, wherein a plurality of items 508 exist within the selected hub group. The user may browse through the items in the various hub groups by selecting different buttons in the hub tab area 502, and may access the item page for the item by selecting the presented item.

[0057] Figure 6 Another user interface 600 for presenting hub groups is presented. Figure 6 In the example of FIG. 6 , user interface 600 presents each hub group, including hub group 602, hub group 604, and hub group 606. Each hub group includes a hub name, a hub description, and the items within the hub group. For example, hub group 602 includes hub name 608, hub description 610, and items presented in item area 612. A user can browse items in various hub groups and access the item page of an item by selecting a presented item.

[0058] Reference now Figure 7 , a flowchart showing an overall method 700 for generating and providing a hub group is provided. The method 700 may be, for example, Figure 1 The method 700 and any other method described herein may be performed by the search / recommendation system 104. Each block of the method 700 and any other method described herein includes a computing process performed using any combination of hardware, firmware, and / or software. For example, various functions may be performed by a processor executing instructions stored in a memory. The method may also be embodied as computer-usable instructions stored on a computer storage medium. These methods may be provided by a standalone application, a service or a hosted service (standalone or in combination with another hosted service), or a plug-in to another product, to name a few.

[0059] As shown in block 702, a hub is generated using a generative model. The hub includes a hub name and a hub description, which include text generated by the generative model. The hub can be generated, for example, by referring to Figure 8 The hub embedding for each hub is generated by the method 800 described above, as shown in block 704. The hub embedding can be generated, for example, by referring to Fig. 9The hub group of the plurality of items is generated using the hub embedding and the item embedding of the item, as shown in block 706. The hub group can be generated, for example, by referring to Fig.10 The hub group is provided to the user device for presentation, as shown in block 708 .

[0060] Figure 8 A flow chart showing a method 800 for generating a hub is presented. The method 800 may be performed, for example, by Figure 1 As shown in block 802, item information of a plurality of items is obtained. For example, the item information can be retrieved from an item data storage device that stores information of items available from a listing platform. In some configurations, the item information of the items is retrieved according to a specific item category.

[0061] As shown in block 804, a prompt is generated using the item information. The prompt may include information for instructing a generative model to generate a hub (including a hub name and a hub description) based on the item information. In some configurations, the prompt includes one or more rules that instruct the generative model how to generate the hub name and the hub description.

[0062] The prompt is provided as input to the generative model, as shown in box 806. Based on the prompt, the generative model generates multiple hubs, as shown in box 808. Specifically, the generative model generates a hub name and text of the hub description for each hub. At box 810, hub post-processing is performed. Hub post-processing can, for example, include: deduplicating similar hubs, confirming the usefulness of the generated hubs, and / or causing the generative model to refine the hubs.

[0063] Fig. 9 A flow chart showing a method 900 for generating a pivot embedding is provided. The method 900 may be, for example, Figure 1 As shown in block 902, items are assigned to hubs. For example, the assignment of items to hubs can be performed by prompting a generative model to perform the assignment. The items can include items used by the generative model to generate hubs (e.g., via method 800) and / or other items sampled from an item data storage device for the purpose of generating a hub embedding.

[0064] As shown in block 904, an item embedding for the item is obtained. In some cases, the item embedding has been previously generated and stored. In this case, the item embedding is retrieved from a storage device. In other cases, the item embedding is generated as part of method 900. The item embedding may include, for example, a text-based embedding generated from a text description of the item or a multimodal embedding (e.g., generated from text and an image of the item).

[0065] A hub embedding is generated from the item embeddings, as shown at block 906. For a given hub, a hub embedding is generated from the item embeddings of the items assigned to the hub at block 902. For example, the hub embedding for the hub may include an average of the item embeddings of the items assigned to the hub.

[0066] Fig.10 A flow chart showing a method 1000 for generating a hub group is provided. The method 1000 may be, for example, Figure 1 1000. As shown in block 1002, item information of an item to be returned to a user device is obtained. The item may be, for example, a recommended item or a search result item. An item embedding of the item is obtained, as shown in block 1004. In some cases, the item embedding has been previously generated and stored. In this case, the item embedding is retrieved from a storage device. In other cases, the item embedding is generated as part of method 1000. The item embedding may, for example, include a text-based embedding or a multimodal embedding (e.g., generated from text and an image of the item) generated from a text description of the item.

[0067] As shown in block 1006, hub information for generating a hub group of items is obtained. The hub information may include hub embeddings for a plurality of hubs. In some configurations, the hub information is for hubs corresponding to a particular category of items corresponding to the grouped items. For example, if the items correspond to a category of shoes, hub information for hubs corresponding to the category of shoes may be obtained.

[0068] Items are assigned to hubs using the hub embeddings and the item embeddings, as shown in block 1008. For example, a similarity metric (such as cosine similarity) can be used to compare the item embeddings and the hub embeddings. A given item can be assigned to a given hub based on a similarity metric of the item embedding to the hub's hub embedding being greater than a similarity metric of the item embedding to other hub embeddings.

[0069] Item and hub rankings are performed, as shown in box 1010. Item rankings for items can be provided based on, for example, the relevance of the items as recommended items or search results. Items can be ranked within each hub group based on the item rankings. A hub group ranking for a hub group can be determined based on the item rankings of the items assigned to each hub group. For example, the hub group ranking for a hub group can be a function (e.g., an average) of the item rankings of the items assigned to the hub group. Hub groups can be ranked relative to each other based on their respective hub rankings.

[0070] Generate hub group information, as shown in box 1012. The hub group information for each hub group may include, for example, a hub name, a hub description, and an ordered sequence of items. The hub group information may further elaborate on the order of the hub groups based on the hub ranking. The hub group information may be provided to a user device to present the hub groups on a user interface.

[0071] Having described the embodiments of the present disclosure, an exemplary operating environment in which embodiments of the present technology can be implemented is described below to provide a general context for various aspects of the present disclosure. Fig.11 , an exemplary operating environment for implementing embodiments of the present technology is shown and generally designated as computing device 1100. Computing device 1100 is merely an example of a suitable computing environment and is not intended to suggest any limitation as to the scope of use or functionality of the technology. Neither should computing device 1100 be interpreted as having any dependency or requirement relating to any one or combination of components illustrated.

[0072] The technology may be described in the general context of computer code or machine-usable instructions, including computer-executable instructions (such as program modules) executed by a computer or other machine (such as a personal data assistant or other handheld device). In general, program modules, including routines, programs, objects, components, data structures, etc., refer to code that performs specific tasks or implements specific abstract data types. The technology may be practiced in a variety of system configurations, including handheld devices, consumer electronics, general-purpose computers, more specialized computing devices, etc. The technology may also be practiced in distributed computing environments, where tasks are performed by remote processing devices that are linked through a communications network.

[0073] refer to Fig.11 , computing device 1100 includes a bus 1110 that directly or indirectly couples the following devices: memory 1112, one or more processors 1114, one or more presentation components 1116, input / output (I / O) ports 1118, input / output components 1120, and an illustrative power supply 1122. Bus 1110 represents what may be one or more buses (e.g., an address bus, a data bus, or a combination thereof). Although for clarity, Fig.11 The various boxes of FIG. 1 are represented by lines, but in reality, it is not so clear to depict the various components, and metaphorically, the lines would more accurately be gray and fuzzy. For example, a presentation component such as a display device can be considered an I / O component. In addition, a processor has memory. The inventors recognize that this is the nature of the art and reiterate that Fig.11 The figures are merely illustrative of exemplary computing devices that can be used in conjunction with one or more embodiments of the present technology. No distinction is made between categories such as "workstation," "server," "laptop," "handheld device," etc., as all of these categories are in the same Fig.11 and refers to “computing devices”.

[0074] The computing device 1100 typically includes a variety of computer-readable media. Computer-readable media can be any available media that can be accessed by the computing device 1100, and includes volatile and non-volatile media, removable and non-removable media. By way of example and not limitation, computer-readable media may include computer storage media and communication media. Computer storage media includes volatile and non-volatile, and removable and non-removable media implemented in any method or technology for storing information, such as computer-readable instructions, data structures, program modules, or other data.

[0075] Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disks (DVD) or other optical disk storage devices, magnetic cassettes, magnetic tape, magnetic disk storage devices or other magnetic storage devices, or any other medium that can be used to store the desired information and can be accessed by the computing device 1100. The terms "computer storage medium" and "computer storage medium" do not themselves include signals.

[0076] Communication media typically embodies computer-readable instructions, data structures, program modules, or other data in the form of a modulated data signal (such as a carrier wave or other transmission mechanism), and includes any information transmission medium. The term "modulated data signal" refers to a signal that is set or changed in a manner that encodes the information in the signal, and the signal has one or more characteristics. By way of example and not limitation, communication media include wired media such as a wired network or a direct wired connection, and wireless media such as acoustic, RF, infrared, and other wireless media. Combinations of any of the above should also be included within the scope of computer-readable media.

[0077] Memory 1112 includes computer storage media in the form of volatile and / or non-volatile memory. Memory may be removable, non-removable, or a combination thereof. Exemplary hardware devices include solid-state memory, hard drives, optical drives, and the like. Computing device 1100 includes one or more processors that read data from various entities such as memory 1112 or I / O components 1120. Presentation component 1116 presents data indications to a user or other device. Exemplary presentation components include display devices, speakers, printing components, vibration components, and the like.

[0078] I / O port 1118 allows computing device 1100 to be logically coupled to other devices including I / O component 1120, some of which may be built-in. Illustrative components include microphones, joysticks, game controllers, satellite antennas, scanners, printers, wireless devices, etc. I / O component 1120 may provide a natural user interface (NUI) that processes air gestures, voice, or other physiological inputs generated by a user. In some cases, the input may be sent to an appropriate network element for further processing. NUI may implement any combination of the following: voice recognition, touch and stylus recognition, facial recognition, biometric recognition, gesture recognition on and near the screen, air gestures, head and eye tracking, and touch recognition associated with a display on computing device 1100. Computing device 1100 may be equipped with a depth camera such as a stereo camera system, an infrared camera system, an RGB camera system, and a combination thereof for gesture detection and recognition. In addition, computing device 1100 may be equipped with an accelerometer or gyroscope capable of detecting motion.

[0079] The present technology has been described with reference to specific embodiments which are intended in all respects to be illustrative rather than restrictive. Alternative embodiments will become apparent to those of ordinary skill in the art to which the present technology pertains without departing from the scope of the present technology.

[0080] Various components used herein have been identified, and it should be understood that any number of components and arrangements can be used to implement the desired functions within the scope of the present disclosure. For example, for the sake of conceptual clarity, the components in the embodiments depicted in the accompanying drawings are shown with lines. Other arrangements of these and other components can also be implemented. For example, although some components are depicted as single components, many elements described herein can be implemented as discrete or distributed components or combined with other components, and implemented in any suitable combination and position. Some elements can be omitted completely. In addition, as described below, the various functions performed by one or more entities described herein can be performed by hardware, firmware and / or software. For example, various functions can be performed by a processor that executes instructions stored in a memory. Therefore, in addition to or in place of the arrangements and elements shown, other arrangements and elements (e.g., machines, interfaces, functions, commands, and functional groupings) can also be used.

[0081] The embodiments described herein may be combined with one or more of the specifically described alternatives. Specifically, the claimed embodiments may include references to more than one other embodiment in the alternatives. The claimed embodiments may specify additional limitations of the claimed subject matter.

[0082] The subject matter of embodiments of the present technology is specifically described herein to meet statutory requirements. However, this description itself is not intended to limit the scope of this patent. Instead, the inventors have contemplated that the claimed subject matter may also be embodied in other ways in conjunction with other prior art or future technologies to include different steps or combinations of steps similar to the steps described in this document. In addition, although the terms "step" and / or "box" may be used herein to refer to different elements of the method employed, the terms should not be interpreted as implying any particular order between or among the various steps disclosed herein unless and in addition to the order of the various steps being explicitly described.

[0083] For purposes of this disclosure, the word "including" has the same broad meaning as the word "comprising," and the word "accessing" includes "receiving," "referencing," or "retrieving." Furthermore, the word "communicating" has the same broad meaning as the word "receiving" or "sending," which is facilitated by a software or hardware-based bus, receiver, or transmitter using the communication media described herein. In addition, unless otherwise indicated, words such as "a," "an," and "an" include the plural as well as the singular. Thus, for example, where there are one or more features, the constraint of "a feature" is satisfied. In addition, the term "or" includes conjunctions, disjunctions, and both (a or b therefore includes a or b, and a and b).

[0084] For the purpose of the above detailed discussion, embodiments of the present technology are described with reference to a distributed computing environment; however, the distributed computing environment described herein is exemplary only. Components may be configured to perform novel embodiments in the embodiments, where the term "configured to" may refer to "programmed to" perform a specific task or implement a specific abstract data type using code. In addition, while embodiments of the present technology may generally refer to the technical solution environments and schematic diagrams described herein, it should be understood that the described technology may be extended to other implementation contexts.

[0085] From the foregoing it will be seen that the technology is well adapted to attain all of the ends and objects set forth above, and other advantages which are obvious and inherent to the system and method. It will be appreciated that some features and subcombinations are useful and may be employed without reference to other features and subcombinations. This is contemplated by the claims and is within the scope of the claims.

Claims

1. One or more computer storage media storing computer usable instructions that, when used by a computing device, cause the computing device to perform operations comprising: accessing item information for a first plurality of items; causing a generative model to generate a plurality of hubs using the item information, each hub in the plurality of hubs comprising a hub name and a hub description; Generate hub embeddings for each hub; obtaining an item embedding for each item in a second plurality of items; assigning items in the second plurality of items to hubs in the plurality of hubs to generate a plurality of hub groups based on the item embeddings and the hub embeddings; as well as User interface information is provided to present at least a portion of the plurality of hub groups on a user device.

2. One or more computer storage media according to claim 1, wherein: Accessing item information for the first plurality of items includes: selecting a first item category from a plurality of item categories; and The first plurality of items is selected based on the first item category.

3. The one or more computer storage media of claim 1, wherein: Causing the generative model to generate the plurality of hubs comprises: generating a prompt comprising item information of the first plurality of items and one or more rules for instructing the generative model to generate a hub name and a hub description for the hub; and The prompt is provided as input to the generative model.

4. The one or more computer storage media of claim 1, wherein: Generating the hub embedding for each hub includes generating a first hub embedding for a first hub of the plurality of hubs by: associating a third plurality of items with the first hub; obtaining an item embedding for each item in the third plurality of items; as well as The first pivot embedding is generated using the item embedding of each item in the third plurality of items.

5. One or more computer storage media according to claim 4, wherein: The third plurality of items includes a subset of one or more items in the first plurality of items, and wherein the generative model associates the third plurality of items with the first hub.

6. The one or more computer storage media of claim 1, wherein: The plurality of hubs are associated with categories of items, and wherein assigning an item in the second plurality of items to a hub in the plurality of hubs comprises: determining that the second plurality of items are associated with the category of items; and The plurality of hubs are selected based on the second plurality of items being associated with the category of items.

7. The one or more computer storage media of claim 1, wherein: The operations also include: determining an item rank for each item in the second plurality of items; and A hub ranking is determined for each hub group in the plurality of hub groups.

8. The one or more computer storage media of claim 7, wherein: A hub rank for a first hub group in the plurality of hub groups is determined based on an item rank assigned to each item in a first subset of items in the first hub group.

9. The one or more computer storage media of claim 7, wherein: The user interface information organizes the hub groups based on the hub rankings.

10. The one or more computer storage media of claim 1, wherein: The second plurality of items includes recommended items or search result items.

11. A computer-implemented method comprising: Access item information for multiple items; generating a prompt including information about the item; providing the prompt as input to a generative model so that the generative model generates text for a plurality of hubs, the text for each hub including a hub name and a hub description; generating a hub embedding for each hub in the plurality of hubs; as well as Hub information of the plurality of hubs is stored on one or more computer storage media, the hub information including the hub name, the hub description and the hub embedding.

12. The computer-implemented method of claim 11, wherein: The method further comprises: Access an item's item embedding; selecting a first hub from the plurality of hubs based on the item embedding and the first hub embedding of the first hub; and An association between the item and the first hub is stored on the one or more computer storage media.

13. The computer-implemented method of claim 11 , wherein: The method further comprises: receiving information identifying a second plurality of items; obtaining an item embedding for each item in the second plurality of items; generating a hub group by assigning items in the second plurality of items to hubs in the plurality of hubs using the item embeddings and the hub embeddings of the plurality of hubs; and User interface information is provided to present at least a portion of the hub group on a user device.

14. The computer-implemented method of claim 13, wherein: The method further comprises: identifying an item category associated with the second plurality of items; and The plurality of hubs are selected based on the item category.

15. The computer-implemented method of claim 13, wherein: The method further comprises: receiving an item rank for each item in the second plurality of items; Determining a hub group ranking of the plurality of hub groups using the item ranking; and The user interface information is generated based at least in part on the item ranking and the hub group ranking.

16. The computer-implemented method of claim 11, wherein: The prompt is generated to include one or more rules that instruct the generative model to generate hub names and hub descriptions for the plurality of hubs.

17. The computer-implemented method of claim 11, wherein: Generating a hub embedding for each hub in the plurality of hubs includes: accessing item information for a second plurality of items; generating a second prompt including the item information for the second plurality of items and the hub names and the hub descriptions for the plurality of hubs; providing the second prompt to the generative model so that the generative model outputs an assignment of items in the second plurality of items to the hubs in the plurality of hubs; obtaining an item embedding for the item in the second plurality of items; and The hub embedding of the plurality of hubs is generated based on the item embedding and the assignment of the items in the second plurality of items to the hubs in the plurality of hubs.

18. A computer system comprising: processor; as well as A computer storage medium storing computer usable instructions that, when used by a processor, cause a computer system to perform operations including: Receive item information for multiple items; determining item categories associated with the plurality of items; selecting a subset of hubs based on the item category, the subset of hubs comprising two or more hubs from a plurality of hubs, each hub from the plurality of hubs comprising a hub name and a hub description generated by a generative model; obtaining an item embedding for each item in the plurality of items; generating a hub group by grouping items from the plurality of items with the hubs from the subset of hubs; as well as User interface information is provided to present at least a portion of the hub group on a user device.

19. The computer system of claim 18, wherein: The plurality of items include recommended items or search result items.

20. The computer system of claim 18, wherein: The operations also include: Obtaining item rankings of the plurality of items; Determining a hub group ranking for the hub group using the item ranking; and The user interface information is generated based on the item ranking and the hub ranking.

Images